My ThinkStation PGX switches itself off while running ordinary workloads. This happens when I transcribe a meeting recording and when I run LLM inference. These are normal jobs that a machine of this class should handle easily. I would like to know if the unit is faulty and what the next steps are.
Setup: upright on an open desk, vents clear, air-conditioned room
Workload: Llama 3.3 70B (AWQ) on vLLM using about 45% of GPU memory, plus WhisperX for transcription
What I observe
High idle temperature. All 7 ACPI thermal zones (acpitz) read 61.7 to 62.8 C when idle. With the model loaded in memory and nothing running, they now read about 66 C. Each zone only exposes a critical trip point of 104.8 C.
Very fast heating. Example from Oct 15, 2026: between 20:28:49 and 20:29:08 the hottest zone went from 69.1 C to 97.8 C in about 20 seconds. The GPU went from 65 C to 87 C, power from 14 W to about 85 W (of 240 W), and the GPU clock dropped from 2444 MHz to 2262 MHz.
Abrupt shutdown. The machine switched off at about 20:29 with no shutdown entry in the journal and no kernel warning. This is well below the 104.8 C critical trip point, so some other protection seems to act first.
Other runs peaked at 96 to 97.4 C at only 75 to 85 W, and the same kind of unexplained shutdown happened on other days too.
The OS shows no fan speed (no fan entries under /sys/class/hwmon), so I cannot check the fans myself.
My questions
Is it normal for this unit to idle at 62 to 66 C and reach about 97 C within 20 seconds at about 85 W?
Is there a newer BIOS or embedded controller firmware than S0QKT0EA, even if it is not listed in LVFS?
How can I check the fans and thermal sensors (diagnostic tool or procedure)? I can run the NVIDIA field diagnostics if you want the result.
Does this unit need repair or replacement? Please tell me the next steps.
I can share the full temperature log (every 2 seconds), the journal of the previous boot, and nvidia-smi output.
I posted this thread on the Lenovo forum because one of my two PGX ThinkStations showed temperatures above 90°C while the other machine remained well below 90°C during identical workloads.
Long story short, some PGX batches have an issue with the thermal paste applied to the CPU and GPU. I filed a warranty claim under Lenovo’s Premiere Program and a technician was sent to my home to repaste the CPU and GPU. Here is a photo of the operation:
Overall, my affected PGX machine runs a few degrees cooler now and I have had no shutdown events since the repasting. And communicating with Lenovo support was an efficient experience, so I can recommend reaching out to their team.
Thanks @kevinpw for the info.
i followed same and their staff has generated ticket, soon technician will visit and resolve the issue.
i agree they really have good customer support
At first, I would let the DGX Dasboard update everything to the latest versions (kernel 7.x, driver 580.178 etc.). At certain points, there were patch versions in which the temperature management was faulty.
Concern Regarding Lenovo ThinkStation PGX Repair Process and Delays
I am sharing my experience with Lenovo Support regarding my ThinkStation PGX and would appreciate some guidance from the community.
My system requires repair, but the concerning part is that Lenovo is arranging replacement parts before a technician has even visited to diagnose the actual problem.
After waiting for an update, I received an email from Lenovo Support apologizing for the delay and informing me that the required parts are being sourced and shipped from China, with an estimated availability date of 5 November 2026.
What I find difficult to understand is how the required parts have been identified without an initial physical diagnosis. No technician has visited to inspect the system or determine the root cause of the problem. It could potentially be something relatively simple that does not require replacing any parts.
I have asked Lenovo to arrange a technician visit first so the actual issue can be identified before proceeding with a parts replacement.
This situation raises a few important questions:
Why are replacement parts being ordered before a technician has physically diagnosed the problem?
Shouldn’t the root cause be identified before determining which components need replacement?
Is this the standard repair procedure for Lenovo ThinkStation PGX systems?
I purchased this workstation to support my AI development and computing workloads, and having it unavailable for an extended period directly affects my work.
